LIVE · PRE-SEEDDaleel MRO
A 29-step retrieval pipeline for aviation maintenance documents. Built so the model stops making things up when a wrong answer could ground a plane.

I build products that actually make it to users. The kind where the Figma file and the live app match up, because I draw one and write the other.
I work on my own. Fullstack plus design. Figma for the interface, React and Next.js for the code, and I'm the one pushing to production at the end of the week. Nothing gets handed between people, so nothing gets lost in the gap between screens and reality.
Most of my time right now goes to AI products. The clearest example is Daleel MRO. It's a 29-step retrieval pipeline for aviation maintenance documents, with an authority hierarchy that keeps manufacturer manuals above operator bulletins, plus a second-pass check before the model is allowed to answer. I wrote the schema, the backend, the evals, and the UI. Nobody else touched it.
Before I went independent, I spent five years as an EASA Part-145 licensed aviation maintenance engineer. In that job, "mostly works" gets someone killed. I carried the instinct over.
LIVE · PRE-SEEDA 29-step retrieval pipeline for aviation maintenance documents. Built so the model stops making things up when a wrong answer could ground a plane.
LIVEMarketing autopilot for real estate. A listing goes in; per-platform social campaigns come out, fair-housing checked, auto-posted, and lead-tracked. Plus a photographer marketplace.
CLIENT · SHIPPEDA video-first luxury real estate platform for Oman, built around reels, AI-assisted browsing, saved comparisons, and automatic Instagram-powered listing updates.
LAUNCHINGAn Arabic-first online plant shop for Oman, priced in Omani Rial, with a full catalog, same-day Muscat delivery, and WhatsApp checkout.
IN PROGRESSA desktop studio that turns Quran recitation audio into finished videos — synchronized Arabic text, verse timing, and a one-click rendering pipeline.
LIVEA trader-grade directory for futures prop firms — real rules, fees and payout terms parsed from each firm's own rulebook, side by side.
IN PROGRESSLive market analysis with an ICT detection engine and Kingline models. Backtesting and a multi-timeframe scanner on top.
LLM work where the model is load-bearing. Retrieval, agent flows, eval harnesses, and the unglamorous work of getting it to stop making things up. I've done this end-to-end for aviation. Your domain is probably easier.
Front to back. Next.js and React for the interface, Python or Node on the server side, Postgres for data. I close features, not just Jira tickets.
Figma prototypes, design tokens, component libraries. They still hold up after handover, because the person writing the React next week is me.
A kickoff call. We cover the problem, who uses it, what breaks you today, what the architecture needs to look like. Anything that's explicitly not in scope also gets written down so nobody guesses later.
Wireframes move to hi-fi in Figma. You get a clickable prototype before I touch a code editor. Once you've signed off on the flow, I start building.
A new build goes up to staging every week. Actual code against actual data, not screenshots in a slide. You see progress each Monday.
Production launch, monitoring set up, docs written, handover done. I stay on retainer for 30 days afterwards to patch whatever breaks and help whoever takes it over.
Ayman does the design and the code, and it doesn't show where one stops and the other starts. Most engineers I've worked with can't pull that off. Neither can most designers.Technical Lead · Regional Aviation